fwiw, criterion.weighted_n_<left/right>can never be 0. Technically it's set to 0 in some functions but attains a positive nonzero value by the end of the function (so nonzero for all intents and purposes)

Thanks @nelson-liu!

Just a note that we're focusing on a release now. If it's quiet around
here, please ping when 0.18 is out.

On 29 August 2016 at 05:46, PatrickOReilly notifications@github.com wrote:

Thanks @nelson-liu https://github.com/nelson-liu!

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
#7266 (comment),
or mute the thread
https://github.com/notifications/unsubscribe-auth/AAEz68He_PmSo3rR7VFM2tMrjuqy35Evks5qkeWUgaJpZM4JuuVb
.

tests are unhappy ;)

Thanks @amueller, I think the tests are better now but the Travis build fails with sklearn.tree._splitter.Splitter has the wrong size, try recompiling. Expected 184, got 176. For me, running python setup.py clean --all before the build resolves this. Is this expected behaviour?

I'm running some tests to see if the runtime is affected when monotonic_constraint == 0. In the meantime, do you have any thoughts on the public API? I've gone with DecisionTreeRegressor(..., increasing=[0,3], decreasing=[2,5]). Currently the arguments must be python lists, perhaps they should rather be 'array-like'?

the test error is because of travis caching. I'll try to fix it. I think you should accept array-likes, i.e. anything that can be converted to an array. You should validate them with the check_array function.

hm for some reason I don't see the cache for this PR on travis :-/

any update on this @amueller? seems like monotonicity pull-requests have been close for over a year but not quite getting merged

This would be a great feature to merge.

@PatrickOReilly are you still working on this?

Would anyone care to pick this up?

I'm planning on having a closer look at this.

@samronsin Just in case, depending on how important it is to support this for random forests, you might want to try implementing this for the new GBDTs instead (#12807)